Proc. Natl. Acad. Sci. USA
Vol. 96, pp. 2758–2763, March 1999
Biochemistry
Toward controlling gene expression at will: Selection and design
of zinc finger domains recognizing each of the 5*-GNN-3* DNA
target sequences
DAVID J. SEGAL, BIRGIT DREIER, ROGER R. BEERLI,
AND
CARLOS F. BARBAS III†
The Skaggs Institute for Chemical Biology and the Department of Molecular Biology, The Scripps Research Institute, La Jolla, CA 92037
Communicated by Paul R. Schimmel, The Scripps Research Institute, La Jolla, CA, December 31, 1998 (received for review October 6, 1998)
zinc finger domain allow for the recognition of longer asymmetric sequences of DNA by this motif.
Recognition of these unique properties led us to propose
and perform experiments aimed at creating what might be a
universal system for the control of gene expression. In recent
experiments we have described polydactyl zinc finger proteins
that contain six zinc finger domains and bind 18 bp of
contiguous DNA sequence (12). Recognition of 18 bp of DNA
is sufficient to describe a unique DNA address within all
known genomes, a requirement for our proposal for using
polydactyl proteins as highly specific gene switches. Indeed, we
have demonstrated control of both gene activation and repression by using these polydactyl proteins in a model system (12).
Because each zinc finger domain typically binds 3 bp of
sequence, a complete recognition alphabet requires the characterization of 64 domains. Existing information, which could
guide the construction of these domains, has come from three
types of studies: structure determination (4–11, 13, 14), sitedirected mutagenesis (15–20), and phage-display selections
(21–27). All have contributed significantly to our understanding of zinc fingeryDNA recognition, but each has its limitations. Structural studies have identified a diverse spectrum of
proteinyDNA interactions but do not explain whether alternative interactions might be more optimal. Further, while
interactions that allow for sequence specific recognition are
observed, little information is provided on how alternate
sequences are excluded from binding. These questions have
been partially addressed by mutagenesis of existing proteins,
but the data are always limited by the number of mutants that
can be characterized. Phage display and selection of randomized libraries overcomes certain numerical limitations, but
providing the appropriate selective pressure to ensure that
both specificity and affinity drive the selection is difficult.
Experimental studies from several laboratories (21–26), including our own (27), have demonstrated that it is possible to
design or select a few members of this recognition alphabet.
However, the specificity and affinity of these domains for their
target DNA was rarely investigated in a rigorous fashion in
these early studies.
In this work we have taken a more systematic approach. We
describe the selection by phage display, refinement by sitedirected mutagenesis, and rigorous characterization of 16 zinc
finger domains representing the 59-GNN-39 subset of this
64-member recognition code. We demonstrate that the identity of the residues at the three helical positions 21, 3, and 6
of a zinc finger domain are typically insufficient to describe in
detail the specificity of the domain. While current zinc finger
recognition codes attempt to define the specificity of the
domain based on the residue identity at helical positions 21,
3, and 6, our results suggest that the predictive value of this
code is limited.
ABSTRACT
We have taken a comprehensive approach to
the generation of novel DNA binding zinc finger domains of
defined specificity. Herein we describe the generation and
characterization of a family of zinc finger domains developed
for the recognition of each of the 16 possible 3-bp DNA binding
sites having the sequence 5*-GNN-3*. Phage display libraries
of zinc finger proteins were created and selected under
conditions that favor enrichment of sequence-specific proteins. Zinc finger domains recognizing a number of sequences
required refinement by site-directed mutagenesis that was
guided by both phage selection data and structural information. In many cases, residues not expected to make basespecific contacts had effects on specificity. A number of these
domains demonstrate exquisite specificity and discriminate
between sequences that differ by a single base with >100-fold
loss in affinity. We conclude that the three helical positions
21, 3, and 6 of a zinc finger domain are insufficient to allow
for the fine specificity of the DNA binding domain to be
predicted. These domains are functionally modular and may
be recombined with one another to create polydactyl proteins
capable of binding 18-bp sequences with subnanomolar affinity. The family of zinc finger domains described here is
sufficient for the construction of 17 million novel proteins that
bind the 5*-(GNN)6-3* family of DNA sequences. These materials and methods should allow for the rapid construction of
novel gene switches and provide the basis for a universal
system for gene control.
The paradigm that the primary mechanism for governing the
expression of genes involves protein switches that bind DNA
in a sequence specific manner was established in 1967 (1).
Since that time diverse structural families of DNA binding
proteins have been described. Despite this wealth of structural
diversity, the Cys2-His2 zinc finger motif constitutes the most
frequently used nucleic acid binding motif in eukaryotes. This
observation is as true for yeast as it is for humans. The
Cys2-His2 zinc finger motif, identified first in the DNA and
RNA binding transcription factor TFIIIA (2), is perhaps the
ideal structural scaffold on which a sequence-specific protein
might be constructed. A single zinc finger domain consists of
approximately 30 aa with a simple bba fold stabilized by
hydrophobic interactions and the chelation of a single zinc ion
(2, 3). Presentation of the a-helix of this domain into the major
groove of DNA allows for sequence-specific base contacts.
Each zinc finger domain typically recognizes 3 bp of DNA
(4–7), though variation in helical presentation can allow for
recognition of a more extended site (8–11). In contrast to most
transcription factors that rely on dimerization of protein
domains for extending protein-DNA contacts to longer DNA
sequences or addresses, simple covalent tandem repeats of the
The publication costs of this article were defrayed in part by page charge
payment. This article must therefore be hereby marked ‘‘advertisement’’ in
accordance with 18 U.S.C. §1734 solely to indicate this fact.
Abbreviation: ZBA, zinc buffer A.
†To whom reprint requests should be addressed at: The Scripps
Research Institute, BCC-515, 10550 North Torrey Pines Road, La
Jolla, CA 92037. e-mail: carlos@scripps.edu.
PNAS is available online at www.pnas.org.
2758
Biochemistry: Segal et al.
MATERIALS AND METHODS
Selection by Phage Display. Construction of zinc-finger
libraries by PCR overlap extension was essentially as described
(27). Growth and precipitation of phage were as described (28,
29), except that ER2537 cells (New England Biolabs) were
used to propagate the phage and 90 mM ZnCl2 was added to
the growth media. Precipitated phage were resuspended in
zinc buffer A (ZBA; 10 mM Tris, pH 7.5y90 mM KCly1 mM
MgCl2y90 mM ZnCl2)y1% BSAy5 mM DTT. Binding reactions [500 ml: ZBAy5 mM DTTy1% Blotto (50 mM TriszHCl,
pH 7.4y100 mM NaCly5% nonfat dry milk, Bio-Rad)y
competitor oligonucleotidesy4 mg sheared herring sperm DNA
(Sigma)y100 ml filtered phage (1013 colony-forming units)]
were incubated for 30 min at room temperature, before the
addition of 72 nM biotinylated hairpin target oligonucleotide.
Incubation continued for 3.5 hr with constant gentle mixing.
Streptavidin-coated magnetic beads (50 ml; Dynal) were
washed twice with 500 ml ZBAy1% BSA, then blocked with
500 ml of ZBAy5% Blottoyantibody-displaying (irrelevant)
phage ('1012 colony-forming units) for '4 hr at room temperature. At the end of the binding period, the blocking
solution was replaced by the binding reaction and incubated 1
hr at room temperature. The beads were washed 10 times over
a 1-hr period with 500 ml of ZBAy5 mM DTTy2% Tween 20,
then once without Tween 20. Bound phage were eluted 30 min
with 10 mgyml of trypsin.
Hairpin target oligonucleotides had the sequence 59-BiotinGGACGCN9N9N9CGCGGGTTTTCCCGCGNNNGCGTCC-39, where NNN was the 3-nt finger-2 target sequence and
N9N9N9 its complement. A similar nonbiotinylated oligonucleotide, in which the target sequence was TGG (compTGG),
was included at 7.2 nM in every round of selection to select
against contaminating parental phage. Two pools of nonbiotinylated oligonucleotides also were used as competitors: one
containing all 64 possible 3-nt targets sequences (compNNN),
the other containing all of the GNN target sequences except
for the current selection target (compGNN). These pools
typically were used as follows: round 1, no compNNN or
compGNN; round 2, 7.2 nM compGNN; round 3, 10.8 nM
compGNN; round 4, 1.8 mM compNNN, 25 nM compGNN;
round 5, 2.7 mM compNNN, 90 nM compGNN; round 6, 2.7
mM compNNN, 250 nM compGNN; round 7, 3.6 mM compNNN, 250 nM compGNN.
Multitarget Specificity Assays. The fragment of pComb3H
(28, 30) phagemid RF DNA containing the zinc-finger coding
sequence was subcloned into a modified pMAL-c2 (New
England Biolabs) bacterial expression vector and transformed
into XL1-Blue (Stratagene). Freezeythaw extracts containing
the overexpressed maltose binding protein-zinc finger fusion
proteins were prepared from isopropyl b-D-thiogalactosideinduced cultures by using the Protein Fusion and Purification
System (New England Biolabs). In 96-well ELISA plates, 0.2
mg of streptavidin (Pierce) was applied to each well for 1 hr at
37°C, then washed twice with water. Biotinylated target oligonucleotide (0.025 mg) was applied similarly. ZBAy3% BSA
was applied for blocking, but the wells were not washed after
incubation. All subsequent incubations were at room temperature. Eight 2-fold serial dilutions of the extracts were applied
in binding buffer (ZBAy1% BSAy5 mM DTTy0.12 mg/ml
sheared herring sperm DNA). The samples were incubated 1
hr, followed by 10 washes with water. Mouse anti-maltose
binding protein mAb (Sigma) in ZBAy1% BSA was applied to
the wells for 30 min, followed by 10 washes with water. Goat
anti-mouse IgG mAb conjugated to alkaline phosphatase
(Sigma) was applied to the wells for 30 min, followed by 10
washes with water. Alkaline phosphatase substrate (Sigma)
was applied, and the OD405 was quantitated with SOFTMAX 2.35
(Molecular Devices).
Proc. Natl. Acad. Sci. USA 96 (1999)
2759
Gel Mobility Shift Assays. Fusion proteins were purified
to .90% homogeneity by using the Protein Fusion and
Purification System (New England Biolabs), except that
ZBAy5 mM DTT was used as the column buffer. Protein
purity and concentration were determined from Coomassie
blue-stained 15% SDSyPAGE gels by comparison to BSA
standards. Target oligonucleotides were labeled at their 59 or
39 ends with [32P] and gel purified. Eleven 3-fold serial
dilutions of protein were incubated in 20 ml of binding
reactions (13 binding buffery10% glyceroly'1 pM target
oligonucleotide) for 3 hr at room temperature, then resolved
on a 5% polyacrlyamide gel in 0.53 TBE buffer (90 mM
Trisy64.6 mM boric acidy2.5 mM EDTA, pH 8.3). Quantitation of dried gels was performed by using a PhosphorImager
and IMAGEQUANT software (Molecular Dynamics), and the KD
was determined by Scatchard analysis.
RESULTS AND DISCUSSION
Library Construction and Selection. As in our previous
studies (27), we have used the murine Cys2-His2 zinc finger
protein Zif268 for construction of phage-display libraries.
Zif268 is structurally the most well-characterized of the zinc
finger proteins (4, 5, 31). DNA recognition in each of the three
zinc finger domains of this protein is mediated by residues in
the N terminus of the a-helix contacting primarily 3 nt on a
single strand of the DNA. The operator binding site for this
three-finger protein is 59-GCGTGGGCG-93 (finger-2 subsite
is underlined). Structural studies of Zif268 and other related
zinc finger-DNA complexes (6–11, 13, 14) have shown that
residues from primarily three positions on the a-helix (21, 3,
and 6) are involved in specific base contacts. Typically, the
residue at position 21 of the a-helix contacts the 39 base of that
finger’s subsite while positions 3 and 6 contact the middle base
and the 59 base, respectively.
To select a family of zinc finger domains recognizing the
59-GNN-39 subset of sequences, we constructed two highly
diverse zinc finger libraries in the phage-display vector
pComb3H (28, 30). Both libraries involved randomization of
residues within the a-helix of finger 2 of C7, a variant of Zif268
(27). The NNK library was constructed by randomization of
positions 21, 1, 2, 3, 5, and 6 by using a condon doping strategy
that allows for all amino acid combinations within 32 condons.
The VNS library was constructed by randomization of positions 22, 21, 1, 2, 3, 5, and 6, which precludes Tyr, Phe, Cys,
and all stop condons in its 24-codon set. The libraries consisted
of 4.4 3 109 and 3.5 3 109 members, respectively, each capable
of recognizing sequences of the 59-GCGNNNGCG-39 type.
The size of the NNK library ensured that it could be surveyed
with 99% confidence while the VNS library was highly diverse
but somewhat incomplete. These libraries are, however, significantly larger than previously reported zinc finger libraries
(21–27). Seven rounds of selection were performed on the zinc
finger displaying-phage with each of the 16 59-GCGGNNGCG-39 biotinylated hairpin DNAs targets by using a
solution binding protocol. Stringency was increased in each
round by the addition of competitor DNA. Sheared DNA was
provided for selection against phage that bound nonspecifically to DNA. Stringent selective pressure for sequence specificity was obtained by providing DNA of the 59GCGNNNGCG-39 type as specific competitors (see Materials
and Methods). Excess DNA of the 59-GCGGNNGCG-39 type
was added to provide even more stringent selection against
binding to DNAs with single or double base changes as
compared with the biotinylated target. Phage binding to the
single biotinylated DNA target sequence were recovered by
using streptavidin-coated beads. In some cases the selection
process was repeated. The finger-2 recognition helices of
several randomly chosen seventh-round clones are shown in
Fig. 1.
2760
Biochemistry: Segal et al.
Proc. Natl. Acad. Sci. USA 96 (1999)
FIG. 1. The finger-2 recognition helices of randomly chosen clones from the seventh round of selection. The selection target site is shown to
the left of each set, followed by the frequency with which each sequence was observed. The helix position of each amino acid is shown at the top,
with positions 21, 3, and 6 shown in bold. Boxed sequences were studied in detail.
Striking conservation of all three of the primary DNA
contact positions (21, 3, and 6) was observed for virtually all
the clones of a given target. Although many of these residues
were observed previously at these positions after selections
with much less complete libraries, the extent of conservation
observed here represents a dramatic improvement over earlier
studies (21–25, 27). Typically, phage selections have shown a
consensus selection in only one or two of these positions. The
greatest sequence variation occurred at the residues in positions 1 and 5, which do not make base contacts in the
Zif268yDNA structure and were expected not to contribute
significantly to recognition (4, 5). Variation in positions 1 and
5 also implied that the conservation in the other positions was
the result of their interaction with the DNA and not simply the
fortuitous amplification of a single clone caused by other
reasons. Conservation of residue identity at position 2 also was
observed. The conservation of position 22 is somewhat artifactual; the NNK library had this residue fixed as serine. This
residue makes contacts with the DNA backbone in the Zif268
structure. Both libraries contained an invariant leucine at
position 4, a critical residue in the hydrophobic core that
stabilizes folding of this domain.
Impressive amino acid conservation was observed for recognition of the same nucleotide in different targets. For
example, Asn in position 3 (Asn3) was virtually always selected
to recognize adenine in the middle position, whether in the
context of GAG, GAA, GAT, or GAC. Gln21 and Arg21 were
always selected to recognize adenine or guanine, respectively,
in the 39 position regardless of context. Amide side chain-based
recognition of adenine by Gln or Asn is well documented in
structural studies as is the Arg guanidinium side chain to
guanine contact with a 39 or 59 guanine (6, 7, 10). More often,
however, two or three amino acids were selected for nucleotide
recognition. His3 or Lys3 (and to a lesser extent, Gly3) were
selected for the recognition of a middle guanine. Ser3 and Ala3
were selected to recognize a middle thymine. Thr3, Asp3, and
Glu3 were selected to recognize a middle cytosine. Asp and
Glu also were selected in position 21 to recognize a 39
cytosine, while Thr21 and Ser21 were selected to recognize a
39 thymine.
Characterization of Finger-2 Proteins. Selected Zif268 variants were subcloned into a bacterial expression vector, and the
proteins were overexpressed (finger-2 proteins, hereafter referred to by the subsite for which they were panned). It is
important to study soluble proteins rather than phage fusions
because it is known that the two may differ significantly in their
binding characteristics (32). The specificity profiles of representative clones are shown in Fig. 2. The proteins were tested
for their ability to recognize each of the 16 59-GNN-39 finger-2
subsites by using a multitarget ELISA assay (Fig. 2, filled bars).
This assay provided an extremely rigorous test for specificity
because there were always six ‘‘nonspecific’’ sites that differed
from the ‘‘specific’’ site by only a single nucleotide out of a 9-nt
target. Many of the phage-selected finger-2 proteins showed
exquisite specificity (for example, Fig. 2 a–e), while others
demonstrated varying degrees of crossreactivity (Fig. 2 f, g, i,
k, m, o, q, and s). Proteins pGCG, pGGT, and pGTT (Fig. 2
u, w, and y) actually bound better to subsites other than those
for which they were selected.
Attempts were made to improve binding specificity by
modifying the recognition helix by using site-directed mutagenesis. Data from our selections and structural information
guided mutant design. More than 100 mutant proteins were
characterized in an effort to expand our understanding of the
rules of recognition. Only the best example for each subsite is
shown in Fig. 2 h, j, l, n, p, r, t, v, x, and z. Although helix
positions 1 and 5 are not expected to play a direct role in DNA
recognition, the best improvements in specificity always involved modifications in these positions. These residues have
been observed to make phosphate backbone contacts, which
contribute to affinity in a nonsequence-specific manner. Removal of nonspecific contacts increases the importance of the
specific contacts to the overall stability of the complex, thereby
enhancing specificity. For example, the specificity of proteins
pGAC, pGAA, and pGAG (Fig. 2 k, m, and o) were improved
simply by replacing atypical, charged residues in positions 1
and 5 with smaller, uncharged residues. Protein pGTT (Fig. 2y)
also was improved by a change in position 5 (Fig. 2z), although
several attempts at selection and mutagenesis failed to identify
a protein that could bind subsite GTT without crossreaction.
Biochemistry: Segal et al.
Proc. Natl. Acad. Sci. USA 96 (1999)
2761
FIG. 2. Multitarget ELISA titration assay for binding specificity. At the top of each graph is the DNA finger-2 target site for which each protein
was selected or designed, and the recognition helix of that protein (positions 22 to 6). Helix positions 21, 3, and 6 are in bold. Proteins modified
by site-directed mutagenesis have the prefix ‘‘m’’ before their DNA target. Columns 1–16 (filled bars) represent target oligos with different finger-2
subsites: 1 5 GGG; 2 5 GGA; 3 5 GGT; 4 5 GGC; 5 5 GAG; 6 5 GAA; 7 5 GAT; 8 5 GAC; 9 5 GTG; 10 5 GTA; 11 5 GTT; 12 5 GTC;
13 5 GCG; 14 5 GCA; 15 5 GCT; 16 5 GCC. Columns 17–20 (empty bars) represent oligonucleotide pools with a unique 59 nucleotide in their
finger-2 subsite: 17 5 GNN; 18 5 ANN; 19 5 TNN; 20 5 CNN. (j) Column 22 5 CGC. (i) Column 22 5 TAC. (bb) Column 22 5 TGG. All data
are background subtracted (column 21 5 no target oligo). The height of each bar represents the average normalized titer from two independent
experiments, with the highest signal normalized to the greatest value in columns 1–16 and 17–20. Error bars represent the deviation from the average.
An arrow indicates the position of the cognate target oligonucleotide.
Another class of modifications involved changes to both binding and nonbinding residues. The crossreactivity of protein
pGGG for the finger-2 subsite GAG (Fig. 2g) was abolished by
the modifications His3–Lys and Thr5–Val (Fig. 2h). It is interesting to note that His3 was unanimously selected during panning to
recognize the middle guanine, although Lys3 provided better
discrimination of A and G. This finding suggests that panning
conditions for this protein may have favored selection by a
parameter such as affinity over that of specificity. Indeed, the
affinity of protein pmGGG for subsite GGG is 15-fold less than
that of pGGG (Table 1). In the Zif268 structure, His3 donates a
hydrogen bond to the N7 of the middle guanine (4, 5). This bond
also could be made with N7 of adenine, and in fact, Zif268 does
not discriminate between G and A in this position (31). Although
this reasoning explains the observed crossreactivity of protein
pGGG, His3 was found to specify only a middle guanine in
proteins pGGA, pmGGC, and pmGGT (Fig. 2 a, j, and x), even
though Lys3 was selected during panning for proteins pGGC and
pGGT. It should be noted that Lys3 also is found in finger 2 of
YY1 and finger 1 of TFIIIA where both fingers recognize binding
sites with a middle G (9, 11, 13). The ability of Lys3 to provide
discrimination against adenine recognition at this position had
2762
Biochemistry: Segal et al.
Table 1.
Affinities of finger-2 proteins
Protein1
Finger-2
helix2
pGGG
pmGGG
SRSDHLTR
SRSDKLVR
pGGA
pmGGT
SQRAHLER
STSGHLVR
pmGGC
pmGAG
SDPGHLVR
SRSDNLVR
pmGAA
pGAT
pmGAC
SQSSNLVR
STSGNLVR
SDPGNLVR
pGTG
pmGTG
SRKDSLVR
SRSDELVR
pGTA
SQSSSLVR
pmGTT
pGTC
STSGSLVR
SDPGALVR
pmGCG
SRSDDLVR
pGCA
SQSGDLRR
pmGCT
pGCC
C7
Zif268
STSGELVR
SDCRDLAR
SRSDHLTT
SRSDHLTT
Finger-2
subsite3
GGG
GGG
GTG
GGA
GGT
GGC
GGC
GAG
GGG
GAA
GAT
GAC
GCC
GTG
GTG
GAG
GTA
GTG
GTT
GTC
GCC
GCG
GAG
GCA
GCT
GCT
GCC
TGG
TGG
Proc. Natl. Acad. Sci. USA 96 (1999)
K D,
nM4
0.4
6
.1,400
3
15
.2,400
40
1
45
0.5
3
3
90
3
15
30
25
.1,000
5
40
.4,400
9
6
2
10
65
80
0.5
10
K D. Proty
K D. Zif268
0.04
0.6
0.3
1.5
4.0
0.1
4.5
0.05
0.3
0.3
9.0
0.3
1.5
3.0
2.5
0.5
4.0
0.9
0.6
0.2
1
6.5
8.0
0.05
1
1Protein
designations are as in Fig. 2.
positions 21, 3, and 6 are shown in bold.
3Altered nucleotides are underlined.
4Values represent at least two independent experiments. The SE was 6
50%.
2Helix
not previously been suggested and is not evident from the
structures of these proteins. In a TFIIIA structure this residue is
involved in contact with a phosphate, not a base (9). The multiple
crossreactivities of protein pGTG (Fig. 2s) were similarly attenuated by modifications Lys1–Ser and Ser3–Glu (Fig. 2t), resulting
in a 5-fold loss in affinity (Table 1). The Ser3–Glu modification
of pmGTG (Fig. 2t) was largely accidental; the intention had been
to create a protein that could recognize the subsite GCG. Glu3
has been shown to be very specific for cytosine in binding site
selection studies of Zif268 (31). No structural studies show an
interaction of Glu3 with the middle thymine, and Glu3 was never
selected to recognize a middle thymine in our study or any others
(21–27). Despite this paucity of predictive data, the Ser3–Glu
modification favored the recognition of a middle thymine over
cytosine (compare Fig. 2 s and t). These examples illustrate the
limitations of relying on previous structures and selection data to
understand the structural elements underlying specificity. It also
should be emphasized that improvements by modifications involving positions 1 and 5 could not have been predicted by
existing ‘‘recognition codes’’ (20, 33–35), which typically consider
only positions 21, 2, 3, and 6. Only by the combination of
selection and site-directed mutagenesis can we begin to fully
understand the intricacies of zinc fingeryDNA recognition.
From the combined selection and mutagenesis data it emerged
that specific recognition of many nucleotides could be best
accomplished by using motifs, rather than a single amino acid. For
example, the best specification of a 39 guanine was achieved by
using the combination of Arg21, Ser1, and Asp2 (the RSD motif).
By using Val5 and Arg6 to specify a 59 guanine, recognition of
subsites GGG, GAG, GTG, and GCG could be accomplished by
using a common helix structure (SRSD-X-LVR) differing only in
the position 3 residue (Lys3 for GGG, Asn3 for GAG, Glu3 for
GTG, and Asp3 for GCG). Similarly, 39 thymine was specified by
using Thr21, Ser1, and Gly2 in the final clones (the TSG motif).
This finding is in stark contrast to the prediction of the code that
Asn21 and Gln21 best recognize 39 thymine (34, 35). Further, a
39 cytosine could be specified by using Asp21, Pro1, and Gly2 (the
DPG motif) except when the subsite was GCC; Pro1 was not
tolerated by this subsite. Specification of a 39 adenine was with
Gln21, Ser1, and Ser2 in two clones (QSS motif). Residues at
positions 1 and 2 of the motifs were studied for each of the 39 bases
and found to provide optimal specificity for a given 39 base as
described here (data not shown).
The multitarget ELISA assays were designed with the
assumption that all of the proteins preferred guanine in the 59
position because all proteins contained Arg6, and this residue
is known from structural studies to contact guanine at this
position (4–11, 13). This interaction was demonstrated here by
using the 59 binding site signature assay (ref. 34; Fig. 2, empty
bars). Each protein was applied to pools of 16 oligonucleotide
targets in which the 59 nucleotide of the finger-2 subsite was
fixed as G, A, T, or C (Fig. 2, columns 17, 18, 19, and 20,
respectively) and the middle and 39 nucleotides were randomized. All proteins (Fig. 2 a–z) preferred the GNN pool with
essentially no crossreactivity. As a control we studied p*GGG
that contains Val6. This recognition helix was reported in
another selection study (22). As seen in Fig. 2aa, Val does not
specify a single base at this position. The crossreactivity of
proteins pGGC and pGAC (Fig. 2 i–l) is an artifact as shown
by the lack of binding to subsites CGC (Fig. 2j, column 22) and
TAC (Fig. 2l, column 22). Target oligonucleotides with a
finger-2 subsite of CCC or TCC were found to create a perfect
GGC or GAC subsite, respectively, on their complementary
strand.
The results of the multitarget ELISA assay were confirmed by
affinity studies of purified proteins (Table 1). In cases where
crossreactivity was minimal in the ELISA assay, a single nucleotide mismatch typically resulted in a greater than 100-fold loss
in affinity. This degree of specificity had yet to be demonstrated
with zinc finger proteins. In general, proteins selected or designed
to bind subsites with G or A in the middle and 39 position had the
highest affinity, followed by those that had only one G or A in the
middle or 39 position, followed by those that contained only T or
C. The former group typically bound their targets with a higher
affinity than Zif268 (10 nM), the latter with somewhat lower
affinity, and almost all of the proteins had an affinity lower than
that of the parental C7 protein. Proteins pGTC, pmGCT, and
pGCC had the lowest affinities (40, 65, and 80 nM, respectively)
and yet were among the most specific (Fig. 2 d, r, and e,
respectively) suggesting that specificity can result not only from
specific protein-DNA contacts, but also from interactions that
exclude all but the correct nucleotide and common backbone
interactions.
Position 2 and Target Site Overlap. Asp2 always was coselected
with Arg21 in all proteins for which the target subsite was GNG.
It is now understood that there are two reasons for this. From
structural studies of Zif268 (4, 5), it is known that Asp2 of finger
2 makes a pair of buttressing hydrogen bonds with Arg21 that
stabilize the Arg21y39 guanine interaction, as well as some
water-mediated contacts. However, the carboxylate of Asp2 also
accepts a hydrogen bond from the N4 of a cytosine that is base
paired to a 59 guanine of the finger-1 subsite. Adenine basepaired to T in this position can make an analogous contact to that
seen with cytosine. This interaction is particularly important
because it extends the recognition subsite of finger 2 from three
nucleotides (GNG) to four [GNG(GyT)] (15, 25, 26). This
phenomenon is referred to as target site overlap and has three
important ramifications. First, Asp2 was favored for selection by
our library when the finger-2 subsite was GNG because our
finger-1 subsite contained a 59 guanine. Second, it may limit the
utility of the libraries used in this study to selection on GNN or
Biochemistry: Segal et al.
TNN finger-2 subsites because finger 3 of these libraries contains
an Asp2, which may help specify the 59 nucleotide of the finger-2
subsite to be G or T. In Zif268 and C7, which have Thr6 in finger
2, Asp2 of finger 3 enforces G or T recognition in the 59 position
(TyG)GG (Fig. 2bb). This interaction also may explain why
previous phage display studies, which all used Zif268-based
libraries, have found selection limited primarily to GNN recognition (21, 23–27). One of these studies stated that 59G recognition is coded by Ser6 and Thr6 (34), yet all of the characterized
finger 2 proteins here use Arg6 for exquisite 59G recognition.
Recognition of 59G by Ser6 and Thr6 proteins is likely an artifact
of target site overlap as seen in Zif268 and C7 and therefore is not
a coded interaction.
Finally, target site overlap potentially limits the use of these
zinc fingers as modular building blocks. From structural data it is
known that there are some zinc fingers in which target site overlap
is quite extensive, such as those in GLI (8) and YY1 (9), and
others that are similar to Zif268 and display only modest overlap.
In our final set of proteins, Asp2 is found in pmGGG, pmGAG,
pmGTG, and pmGCG. The overlap potential of other residues
found at position 2 is largely unknown; however, structural studies
reveal that many other residues found at this position may
participate in such cross-subsite contacts. Fingers containing
Asp2 may limit modularity, because they would require that each
GNG subsite be followed by a T or G.
CONCLUSIONS
We have demonstrated that many of the 16 possible GNN triplet
sequences can be recognized with exquisite specificity by zinc
finger domains. Optimized zinc finger domains can discriminate
single base differences by greater than 100-fold loss in affinity.
While many of the amino acids found in the optimized proteins
at the key contact positions 21, 3, and 6 are those that are
consistent with a simple code of recognition, we have discovered
that optimal specific recognition is sensitive to the context in
which these residues are presented. Residues at positions 1, 2, and
5 have been found to be critical for specific recognition. Further
we demonstrate, that in contrast to the expectations of a simple
recognition code, that sequence motifs at positions 21, 1, and 2
rather than the simple identity of the position 1 residue are
required for highly specific recognition of the 39 base. We believe
these residues provide the proper stereochemical context for
interactions of the helix both in terms of recognition of specific
bases and in the exclusion of other bases, the net result being
highly specific interactions. Thus our understanding of a recognition code is weak even when the recognition helix is constrained
within the same zinc finger framework. We anticipate that
attempts to apply a recognition code derived from the study of
finger-2 variants of Zif268 will be limited as the effects of the zinc
finger framework on helix presentation are not appreciated. One
motivation for increasing our understanding of the recognition
codes is to apply it to the many naturally occurring zinc finger
proteins of unknown function. It is clear, however, that many
more studies will be required to make this goal feasible.
Broad utility of the domains described here would be realized
if they were modular in both their interactions with DNA and
other zinc finger domains. This cooperativity could be achieved
by working within the likely limitations imposed by target site
overlap, namely that sequences of the 59-(GNN)x-39 type should
be targeted. Indeed, we have now demonstrated the functional
modularity of the zinc finger domains described here in the
construction of polydactyl proteins that bind 18 bp of DNA with
subnanomolar affinity (36). These polydactyl proteins have been
used to activate and repress transcription driven by the human
erbB-2 promoter in living cells. The family of zinc finger domains
described here should be sufficient for the construction of 166 or
17 million novel proteins that bind the 59-(GNN)6-39 family of
DNA sequences. Together, the materials and methods of these
reports should allow for the rapid construction of novel gene
Proc. Natl. Acad. Sci. USA 96 (1999)
2763
switches and provide the basis for a universal system for gene
control.
We thank Jayant Ghiara for his contributions and Jessica Saldana,
Kris Bower, and Marikka Elia for their technical assistance. This study
was supported in part by National Institutes of Health Grant GM
53910 to C.F.B. Postdoctoral fellowships were received by B.D. from
the Deutsche Forschungsgemeinschaft, and by R.R.B. from the Swiss
National Science Foundation and the Krebsliga beider Basel.
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
Ptashne, M. (1967) Nature (London) 214, 232–234.
Miller, J., McLachlan, A. D. & Klug, A. (1985) EMBO J. 4,
1609–1614.
Lee, M. S., Gippert, G. P., Soman, K. V., Case, D. A. & Wright,
P. E. (1989) Science 245, 635–637.
Pavletich, N. P. & Pabo, C. O. (1991) Science 252, 809–817.
Elrod-Erickson, M., Rould, M. A., Nekludova, L. & Pabo, C. O.
(1996) Structure (London) 4, 1171–1180.
Elrod-Erickson, M., Benson, T. E. & Pabo, C. O. (1998) Structure
(London) 6, 451–464.
Kim, C. A. & Berg, J. M. (1996) Nat. Struct. Biol. 3, 940–945.
Pavletich, N. P. & Pabo, C. O. (1993) Science 261, 1701–1707.
Houbaviy, H. B., Usheva, A., Shenk, T. & Burley, S. K. (1996)
Proc. Natl. Acad. Sci. USA 93, 13577–13582.
Fairall, L., Schwabe, J. W. R., Chapman, L., Finch, J. T. &
Rhodes, D. (1993) Nature (London) 366, 483–487.
Wuttke, D. S., Foster, M. P., Case, D. A., Gottesfeld, J. M. &
Wright, P. E. (1997) J. Mol. Biol. 273, 183–206.
Liu, Q., Segal, D. J., Ghiara, J. B. & Barbas III, C. F. (1997) Proc.
Natl. Acad. Sci. USA 94, 5525–5530.
Nolte, R. T., Conlin, R. M., Harrison, S. C. & Brown, R. S. (1998)
Proc. Natl. Acad. Sci. USA 95, 2938–2943.
Narayan, V. A., Kriwacki, R. W. & Caradonna, J. P. (1997) J. Biol.
Chem. 272, 7801–7809.
Isalan, M., Choo, Y. & Klug, A. (1997) Proc. Natl. Acad. Sci. USA
94, 5617–5621.
Nardelli, J., Gibson, T. J., Vesque, C. & Charnay, P. (1991) Nature
(London) 349, 175–178.
Nardelli, J., Gibson, T. & Charnay, P. (1992) Nucleic Acids Res.
20, 4137–4144.
Taylor, W. E., Suruki, H. K., Lin, A. H. T., Naraghi-Arani, P.,
Igarashi, R. Y., Younessian, M., Katkus, P. & Vo, N. V. (1995)
Biochemistry 34, 3222–3230.
Desjarlais, J. R. & Berg, J. M. (1992) Proteins Struct. Funct.
Genet. 12, 101–104.
Desjarlais, J. R. & Berg, J. M. (1992) Proc. Natl. Acad. Sci. USA
89, 7345–7349.
Choo, Y. & Klug, A. (1994) Proc. Natl. Acad. Sci. USA 91,
11163–11167.
Greisman, H. A. & Pabo, C. O. (1997) Science 275, 657–661.
Rebar, E. J. & Pabo, C. O. (1994) Science 263, 671–673.
Jamieson, A. C., Kim, S.-H. & Wells, J. A. (1994) Biochemistry 33,
5689–5695.
Jamieson, A. C., Wang, H. & Kim, S.-H. (1996) Proc. Natl. Acad.
Sci. USA 93, 12834–12839.
Isalan, M., Klug, A. & Choo, Y. (1998) Biochemistry 37, 12026–
12033.
Wu, H., Yang, W.-P. & Barbas III, C. F. (1995) Proc. Natl. Acad.
Sci. USA 92, 344–348.
Barbas III, C. F., Kang, A. S., Lerner, R. A. & Benkovic, S. J.
(1991) Proc. Natl. Acad. Sci. USA 88, 7978–7982.
Barbas III, C. F. & Lerner, R. A. (1991) Methods Companion
Methods Enzymol. 2, 119–124.
Rader, C. & Barbas III, C. F. (1997) Curr. Opin. Biotechnol. 8,
503–508.
Swirnoff, A. H. & Milbrandt, J. (1995) Mol. Cell. Biol. 15,
2275–2287.
Crameri, A., Cwirla, S. & Stemmer, W. P. (1996) Nat. Med. 2,
100–102.
Suzuki, M., Gerstein, M. & Yagi, N. (1994) Nucleic Acids Res. 22,
3397–3405.
Choo, Y. & Klug, A. (1994) Proc. Natl. Acad. Sci. USA 91,
11168–11172.
Choo, Y. & Klug, A. (1997) Curr. Opin. Struct. Biol. 7, 117–125.
Beerli, R. R., Segal, D. J., Dreier, B. & Barbas, C. F. III (1998)
Proc. Natl. Acad. Sci. USA 95, 14628–14633.